The Biomaterials Annotator: a system for ontology-based concept annotation of biomaterials text - Association for Computational ...
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
The Biomaterials Annotator: a system for ontology-based concept
annotation of biomaterials text
Javier Corvi1 , Carla V. Fuenteslópez2 , José M. Fernández1 , Josep Lluis Gelpí1,3 ,
Maria Pau Ginebra4 , Salvador Capella-Gutierrez1 and Osnat Hakimi1,5
1
Barcelona Supercomputing Center (BSC), Barcelona, Spain
2
Institute of Biomedical Engineering, Botnar Research Centre, University of Oxford, UK
3
Dept. of Biochemistry and Molecular Biology, University of Barcelona, Spain
4
Dept. of Material Science and Engineering, Universitat Politècnica de Catalunya, Spain
5
Faculty of Medicine and Health Sciences, Universitat Internacional de Catalunya, Spain
javier.corvi@bsc.es
osnat.hakimi@gmail.com
Abstract generated by the field is primarily available in
text documents, such as peer-reviewed research
Biomaterials are synthetic or natural materials
papers, patents and conference abstracts. This ever-
used for constructing artificial organs, fabricat-
ing prostheses, or replacing tissues. The last growing knowledge is increasingly harder for re-
century saw the development of thousands of searchers to efficiently discover, organize and use.
novel biomaterials and, as a result, an expo- For example, systematically reviewing the appli-
nential increase in scientific publications in the cations and scaffolds made of a commonly used
field. Large-scale analysis of biomaterials and polymer such as poly-lactic-glycolic-acid (PLGA),
their performance could enable data-driven requires skimming through >12,000+ abstracts
material selection and implant design. How-
(MEDLINE search on October 2020). Among the
ever, such analysis requires identification and
organization of concepts, such as materials and different alternatives for the automated processing
structures, from published texts. To facilitate of available texts, Natural Language Processing
future information extraction and the applica- (NLP) workflows for information retrieval and in-
tion of machine-learning techniques, we de- dexing offer a much needed automated solution.
veloped a semantic annotator specifically tai- Such computational workflows facilitate informa-
lored for the biomaterials literature. The Bio- tion discovery, information extraction and orga-
materials Annotator has been implemented fol- nization, saving researchers time and minimizing
lowing a modular organization using software
manual tasks.
containers for the different components and or-
chestrated using Nextflow as workflow man- Central to information retrieval and indexing is
ager. Natural language processing (NLP) com- the extraction of concepts of interest, also known
ponents are mainly developed in Java. This as Named Entity Recognition (NER). NER is an
set-up has allowed named entity recognition integral part of NLP workflows as it allows the au-
of seventeen classes relevant to the biomateri- tomated identification of concepts in unstructured
als domain. Here we detail the development,
text and its assignment to a pre-defined category
evaluation and performance of the system, as
well as the release of the first collection of an-
or class. For example, in the field of biomaterials,
notated biomaterials abstracts. We make both categories may include ‘Biomaterials’ (‘PLGA’),
the corpus and system available to the commu- ‘Structures’ (such as ‘fibre’ or ‘sponge’) and ‘Tis-
nity to promote future efforts in the field and sues’ (such as ‘tendon’ or ‘bone’). The use of NER
contribute towards its sustainability. to automatically recognize entities enables several
downstream applications, including machine trans-
1 Introduction lation, information retrieval and indexing as well
The last two decades saw the field of biomate- as automated question-answering mechanisms.
rials and tissue engineering grow from a small The recognition of concepts in the biomaterials
niche of biomedical research to an extensive do- domain is complicated by language and terminol-
main, covering topics such as functional materi- ogy originating from multiple scientific disciplines
als, cell-material interaction, nanomaterials and (chemistry, engineering, biology, medicine). A sig-
medical devices. The expanding scientific data nificant challenge lies in identifying and combining
36
Proceedings of the Second Workshop on Scholarly Document Processing, pages 36–48
June 10, 2021. ©2021 Association for Computational LinguisticsFigure 1: Overview of the workflow used in the development and validation of the Biomaterials Annotator.
lexical and semantic resources across domains, and materials Annotator, along with an annotated
thus to date there are no automatic biomaterials- collection of biomaterials literature, are publicly
specific NER systems to detect relevant entities of available for use and further development at
interest. https://github.com/ProjectDebbie/
Here, we report the development of the first Biomaterials_annotator.
biomaterials-specific annotation system, designed
to recognize named entities from seventeen differ- 2 Previous relevant work
ent categories, reflecting the complexity and diver-
sity of contemporary biomaterials research. When Unlike general purpose NLP systems, biomedical
considering approaches for the design of the Bio- domain-specific tools require advanced approaches
materials Annotator, i.e. lexical versus machine to detect classes of interest such as diseases and
learning-based NER, such as CRF or RNN, it was gene names. In this area, there are several well-
essential to consider the number of desired annota- known and widely used systems and tools, such
tion categories in the system (17) and the absence as Metamap (Aronson, 2001) and Pubtator (Wei
of an annotated corpora for text mining efforts for et al., 2013), which were developed using dif-
the majority of them. Based on these premises, it ferent NER methodologies and approaches, e.g.
was concluded that training a model for each cat- gazetteers and hand-made rule-based NER; ma-
egory was impractical. Thus, the system relies on chine learning-based NER that includes Hidden
manually curated and validated lexical resources. Markov Model, Conditional Random Fields (CRF)
and recurrent neural network (RNN); and Hybrid
To cover entities from different domains, mul- NER (Lee et al., 2003; McCallum and Li, 2003;
tiple nomenclatures, vocabularies, and especially Song et al., 2004; GuoDong and Jian, 2004; Zhao,
ontologies were identified and combined. To com- 2004; Yeh et al., 2005; Campos et al., 2013; Song
bine these resources into a single instrument, the et al., 2018; Dang et al., 2018; Kaewphan et al.,
Devices, Experimental scaffolds and Biomateri- 2018; Cho and Lee, 2019). In the context of this
als Ontology (DEB) was used providing the log- work, generic text mining tools previously devel-
ical schema and the definition of key categories oped for the eTRANSAFE project (Pognan et al.,
(Hakimi et al., 2020). 2021) have been adapted and further developed for
The resulting open source-system, the Bio- the biomaterials domain.
37Whilst there are a handful of ontologies in the
biomaterials domain (such the nanoparticle ontol-
ogy NPO (Thomas et al., 2011) and the Bone and
Cartilage Tissue Engineering Ontology BCTEO
(Viti et al., 2014)), to the best of the our knowledge,
the DEB ontology (Hakimi et al., 2020) is the only
one that is tailored to link and curate concepts for
Biomaterials NER. Therefore, it specifically cov-
ers different categories related to the biomaterials
domain.
3 Methodology overview
To develop the annotation tool, the workflow
in Figure 1 was followed. Various corpora of
abstracts were used during the development, Figure 2: Overview of the components of the Bioma-
terials Annotator; including the standard preprocessing
covering the general biomaterials literature.
steps (A) and the biomaterials named entity recognition
These corpora included a collection of man- steps (B).
ually curated abstracts of the biomedical
polymer polydioxanone (Fuenteslópez et al 2021,
manuscript in preparation, GitHub repository: includes the steps previously outlined. This
https://github.com/ProjectDebbie/ component is written in JAVA and it uses the
polydioxanone_project) and a previously Stanford CoreNLP Natural Language Processing
published biomaterials gold standard collection open source toolkit. The use of the Stanford
(Hakimi et al., 2020), comprising a total of 1222 CoreNLP API benefits greatly from the provision
abstracts. Corpora were passed through four steps, of a set of stable, robust, high quality linguistic
each described in detail below. The first step was a analysis components, which can be easily invoked
text preprocessing component (section 3.1). This for common scenarios (Manning et al., 2014).
was followed by concept recognition (section 3.2), The Standard NLP preprocessing component
initially using the MeSH controlled vocabulary is available at https://gitlab.bsc.es/
and the DEB ontology. Then, the annotations inb/text-mining/generic-tools/
were evaluated by two domain experts, errors nlp-standard-preprocessing.
were flagged up and additional lexical resources
were added through keyword searches. Concept 3.2 Concept recognition
recognition, manual evaluation and curation of Here, we developed NER components to detect
lexical resources were performed in an iterative relevant entities related to the biomaterials domain
manner during the development phase (section 3.3) based on the DEB ontology in conjunction with
over 1000 abstracts. Once the development phase other open relevant resources, such as the National
was completed, validation by domain experts was Cancer Institute Thesaurus (NCIT) and (CHEBI).
performed on 199 independent abstracts which A comprehensive description of the resources
were not used during the development process included in this work is described in section 3.3 and
(section 4.1). The resulting annotated collection Appendix B. Lexical resources were transformed
of biomaterials abstracts was published as open into gazetteers to be used in the NER process
source. (Figure 2.B). Internally, the NER process was
divided into three main steps; the MSH Annotator,
3.1 Text preprocessing which annotates relevant categories from the
To prepare the text for concept recognition, MeSH terminology; the Dictionary Annotator,
several Natural Language Processing (NLP) steps which annotated predefined categories from the
were performed, namely: tokenization, sentence relevant dictionaries; and the Post-processing step
splitting, part-of-speech tagging and morpho- in which specific rules were executed. These in-
logical analysis (Figure 2.A). We developed the clude entity recognition based on lexical rules and
Standard NLP preprocessing component which the removal of false positives, among other tasks.
38The MSH Annotator is available at https:// detect concepts:
github.com/ProjectDebbie/debbie_ (Token.pos == "JJ" | Token.pos == "NN") To-
umls_annotations; and the Dictionary An- ken.root == "cell"
notator and Post-processing rules are available at The inclusion of this rule enables the detection of
https://github.com/ProjectDebbie/ Cell-type concepts that are not present in the dic-
DEBBIE_dictionaries_annotations. tionaries; e.g. “neuronal cells”, “cancer cell” and
These components are instances of the nlp-gate- “osteogenic cells”. The discovery of such rules is
generic-component (https://gitlab.bsc. a continuous work; future Biomaterials Annotator
es/inb/text-mining/generic-tools/ versions will improve the lexical rules included to
nlp-gate-generic-component), a detect relevant concepts.
generic component developed in JAVA by our Another key problem to address is the recogni-
team that uses the General Architecture of Text tion of abbreviation concepts; to achieve this prob-
Engineering (GATE) software (Cunningham et al., lem we developed a post-processing rule based
2013) and can be parametrized with gazetteers and on a modified version of Schwartz’s algorithm
specific handmade JAPE (Java Annotation Patterns (Schwartz and Hearst, 2003). First, we detect
Engine) rules. Using the Biomaterials Annotator, an abbreviation candidate given a text pattern
every recognised entity is labelled with one of the (regex=”(?:[a-z]*[A-Z][a-z]*)2,”); subsequently,
categories (Figure 3.A-B). the Schwartz’s algorithm is applied to detect
The nlp-gate-generic-component was configured whether there is a definition that matches the abbre-
to use the GATE Flexible gazetteer, allowing to viation candidate in the sentence; in such case, if
capture the words present in the text as well as the definition has an entity class assigned to it, we
their morphological root value (lemma). This en- annotate the abbreviation with the same class. As
sures that inflected forms of a word (i.e. plural, an example, in the following sentence: “We investi-
singular, -ing forms, tense) can be recognised and gated the potential of human bone marrow derived
analysed as a single item. In addition, the dictio- Mesenchymal stem cells (MSCs) for neuronal differ-
naries used in the Biomaterials Annotator include entiation in vitro....”; the expression ‘Mesenchymal
preferred synonyms, providing the possibility to stem cells’ is annotated under the ’Cell’ category.
map terms semantically to a specific primary con- But the ‘MSCs’ abbreviation is not; moreover in the
cept. Thus, the Biomaterials Annotator performs rest of the text the abbreviation is used instead of
semantic mapping of the annotations by, not only its long form. The abbreviation-rule detects ‘MSCs’
recognizing the category of an entity, but also link- as an abbreviation of ‘Mesenchymal stem cells’ and
ing it to the appropriate entry in a well-established assigns the ’Cell’ category to all the ‘MSCs’ men-
resource (Jovanović and Bagheri, 2017). For exam- tions in the text.
ple, the terms: “canine”, “dogs” and “dog” were
all annotated under the ‘Species’ category; and in- 3.3 Terminologies and ontologies curation
side the features of each annotation the preferred and manual evaluation
term is “dog”. This enables the retrieval of all the
One of the main hurdles to biomaterials concept
corresponding terms using the single search term
recognition is the interdisciplinary nature of the
‘dog’.
domain, with scientific texts containing concepts
To complete the annotation process, the annota- from various fields such as biology, chemistry,
tor executes JAPE rules for post-processing func- engineering and medicine. A key objective of
tions, such as the removal of false positives and the the Biomaterials Annotator was to identify and
addition of information to each annotation. Added combine lexical resources from the different
information includes the ontology source, the ontol- domains in order to cover as many relevant
ogy term id, the lemma and the preferred synonym biomaterials concepts as possible. Resources were
(Figure 3.C-D). In addition, JAPE rules were run identified using a manual, bottom-up approach,
to identify entities using lexical constraints and with cyclic re-iteration, as shown in Figure 1. As a
address the concept recognition of abbreviations. starting point, abstracts were annotated with the
Rule-based entities recognition can use part-of- automated NER approach described in section 3.2
speech of concepts, as an example; in the case using the DEB ontology. After each annotation
of ‘Cell’ category, there is a lexical rule defined to round, manual evaluation was performed by
39Figure 3: The appearance of an annotated abstract on GATE’s user interface. A) Shows the annotated text and in B)
colored labels used to tag annotations by their respective category. C) Information regarding each annotation (type,
position, features), and in D) a specific example: “polymers”: "BiomaterialType" entity with their corresponding
features.
ANNOTATION CATEGORIES
BIOLOGICALLY
RESEARCH
BIOMATERIAL ACTIVE TISSUE CELL TYPE
SUBSTANCE TECHNIQUE
SCANNING
SILK
EXAMPLE: RGD PANCREAS ELECTRON FIBROBLAST
FIBROIN
MICROSCOPY
DEB DEB MESH MESH MESH
MAIN
SEMANTIC MESH MESH UBERON NCIT NCIT
RESOURCES:
CHEBI NCIT PREMEDONTO NPO UBERON
Figure 4: An illustration of part of the annotation schema (showing five out of seventeen categories), which relies
on multiple semantic resources for each annotation category. Full details of all categories and resources are in the
Appendix A.
40two domain experts. The evaluation entailed tion set, resulting from the execution of the Bio-
reviewing samples of 10-20 abstracts in order materials Annotator, was manually verified by 9
to flag annotation errors and highlight relevant biomaterials experts. The validation process was
concepts which were missed by the system. The performed using the GATE user interface, where
flagged terms were used for keyword search in annotations made by the system were presented
the Bioportal (Martínez-Romero et al., 2017) and to the biomaterials experts with the possibility of
the UMLS metathesaurus browser (Bodenreider, adding missing annotations, removing false annota-
2004). Through these searches, specific classes tions and editing annotations. Once the expert had
(or ‘parent concepts’) within relevant ontologies finished the validation of a document, it was saved
and UMLS ‘semantic types’ were identified as a different validated copy.
and added to the annotation schema (part of Two strategies to indicate if two annotations
which is shown for illustration in Figure 4). The agree or not were considered; a strict approach,
resources identified belonged to three categories: in which the annotations agree if they have the
ontologies, controlled vocabularies and nomen- same origin and end offset, and a more relaxed or
clatures. All the ontologies were open access and “lenient” approach, where the annotations agree
downloaded in .owl format from NCBO Bioportal if they overlap at some point. For example, in
(http://bioportal.bioontology.org) the partial approach the biomaterials expressions
(Martínez-Romero et al., 2017). The controlled “polyvinyl alcohol” and “polyvinyl” are considered
vocabulary (MeSH) was downloaded for use to agree, which does not happen in the strict agree-
under license from the UMLS Terminology ment.
Services. The GMDN nomenclature was kindly
To measure the performance of the NER system,
provided in .xml format by the GMDN agency
the set validated by the experts was taken as the
under a license. A summary of all the used
gold standard and the system’s output as the set
resources is in Appendix B. For the resources
to be validated. Table 1 shows the recall, preci-
to be used in the annotation system, relevant
sion and F-score, including the strict and lenient
classes were imported into a dictionary (gazetteer)
approaches, as well as an average between them.
containing the following fields: the term, its
The global scores calculated for the system are also
label (annotation category), the ID and whenever
presented, obtaining an 0.75 strict F-score, 0.79
available, a preferred synonym. The extraction of
lenient F-Score and 0.77 average F-score.
desired classes from the ontologies to dictionary
format was done using an implementation of Figure 5 shows the average F-scores calculated
owlready2 (Lamy, 2017) and the code (named for the different categories. Categories with an av-
owl2dict_light) is available in an open github erage F-score above 0.8 are considered categories
repository as part of the (https://github. in which the concepts are satisfactorily covered by
com/ProjectDebbie/OWL2DICT). The the resources used (e.g. Structure, BiomaterialType
resulting dictionaries are also available and Tissue). On the other hand, there are categories
(https://github.com/ProjectDebbie/ with lower scores, and specifically: ’Biomaterial’,
DEBBIE_dictionaries_annotations). ’Biologically active substance’ and ’Cell’. The cat-
These were in turn used by the Dictionary egories Biomaterials and Biologically active sub-
Annotator component for concept recognition as stance had significantly reduced accuracy because
described above in section 3.2. they include many ambiguous concepts. Some ma-
terials may act as a biomaterial in one set-up, but
4 Results can also be measured in terms of cell expression
or non-biomaterial use in another set-up (e.g. col-
4.1 Expert validation lagen). In the latter case, the human validator will
To measure the efficiency of a text mining system delete the ‘Biomaterial’ annotation. Solving this
such as the Biomaterials Annotator, it is fundamen- kind of ambiguities will require other strategies,
tal to organize and plan a validation stage aimed at such as specific lexical rules or machine learning
indicating the performance of the system. The Bio- approaches. Another factor impeding good quality
materials Annotator was validated through manual annotations of Biomaterials is the lack of good qual-
verification of the validation set, an independent ity vocabulary of medical polymers. Polymer and
collection of 199 abstracts. The annotated valida- co-polymer naming is notoriously variable, with
41Precision Recall F-score Precision Recall F-score Precision Recall F-score
Category
- strict - strict - strict - lenient - lenient - lenient - average - average - average
Adverse Effects 0.94 0.75 0.82 1 0.8 0.87 0.97 0.77 0.85
Associated Biological Process 0.88 0.68 0.77 0.94 0.73 0.82 0.91 0.71 0.79
Biologically Active Substance 0.58 0.43 0.49 0.7 0.52 0.59 0.64 0.48 0.54
Biomaterial 0.76 0.47 0.57 0.83 0.52 0.63 0.79 0.49 0.6
Biomaterial Type 0.92 0.88 0.9 0.98 0.93 0.95 0.95 0.9 0.92
Cell 0.76 0.59 0.66 0.84 0.65 0.73 0.8 0.62 0.69
Effect On Biological System 0.96 0.69 0.79 1 0.72 0.82 0.98 0.71 0.8
Manufactured Object 0.96 0.86 0.9 0.96 0.86 0.9 0.96 0.86 0.9
Manufactured Object Component 0.91 0.84 0.86 0.91 0.84 0.87 0.91 0.84 0.87
Manufactured Object Features 0.68 0.59 0.62 0.71 0.61 0.65 0.69 0.6 0.64
Material Processing 0.78 0.6 0.67 0.83 0.63 0.71 0.81 0.61 0.69
Medical Application 0.68 0.49 0.57 0.82 0.6 0.69 0.75 0.54 0.63
Research Technique 0.81 0.63 0.71 0.87 0.68 0.76 0.84 0.66 0.73
Species 0.97 0.79 0.87 0.99 0.81 0.89 0.98 0.8 0.88
Structure 0.93 0.77 0.84 0.95 0.79 0.86 0.94 0.78 0.85
Study Type 0.96 0.95 0.96 0.99 0.97 0.98 0.98 0.96 0.97
Tissue 0.8 0.77 0.78 0.85 0.82 0.83 0.82 0.8 0.81
Global 0.84 0.69 0.75 0.89 0.73 0.79 0.86 0.71 0.77
Table 1: The performance of the Biomaterials Annotator in a test set of 199 abstracts validated manually by 9
experts.
some named by their commercial name or abbrevi- notator and deposits them in an open access
ation. To address these inaccuracies, future work database. DEBBIE is under development and
will involve expanding relevant ontologies using can be accessed at https://github.com/
tools such as Spike (Taub-Tabib et al., 2020), in- ProjectDebbie/DEBBIE_pipeline.
cluding additional lexical rules, and adding ma-
Category Count
chine learning components. Adverse Effects 657
Associated Biological Process 6231
4.2 Full system implementation and Biologically Active Substance 7709
availability Biomaterial 5726
Biomaterial Type 1543
A significant challenge for scientific software ap- Cell 6839
Effect On Biological System 972
plications is providing facilities to share, distribute Manufactured Object 5967
and run such systems in a simple and convenient Manufactured Object Component 2307
way. Furthermore, an important concern is the Manufactured Object Features 4200
Material Processing 2728
possibility of replicating the results obtained Medical Application 3868
during the research. In order to accomplish these Research Technique 3701
requirements and follow good practices, we de- Species 2089
Structure 4136
veloped the Biomaterials Annotator using Docker Study Type 1806
as software container technology and Nextflow Tissue 9997
as the workflow manager. Through the use of Entities 70476
Tokens 392605
Docker, all the subcomponents of the Biomaterials Sentences 15979
Annotator were individually compartmentalized; Abstracts 1222
hosting their own dependencies and programs that
work only inside the isolated container. In addition, Table 2: Annotated biomaterials corpus statistics.
the Nextflow workflow manager was used for
the automated orchestration and execution of the
pipeline. By using this architecture, the entire tool, 4.3 Annotated corpus release
or any of its individual components, can be easily Another key objective was to generate the first an-
installed and run in heterogeneous environments. notated corpus with entities related to the bioma-
The Biomaterials Annotator is available at terials domain. Such a corpus will facilitate the
https://github.com/ProjectDebbie/ development and evaluation of text mining mod-
Biomaterials_annotator. els for automated extraction of biomaterials-related
The Biomaterials Annotator is part of DEB- data from text.
BIE (Database of biomedical materials), a The biomaterials annotated dataset consists of
wider system that retrieves abstracts from 1222 biomaterials abstracts describing the evalu-
pubmed, annotates using the Biomaterials An- ation of biomaterials in either a laboratory or a
42Figure 5: Average F-score of the automated annotations across categories.
clinical setting. Each abstract is individually con- manner aiming to enhanced performance in key
tained as a separate file under the GATE format. categories such as Biomaterials and Cells. In addi-
Table 2 shows statistics concerning the number of tion, future versions of the Biomaterials Annotator
concepts corresponding to the different categories, will be closely related to the DEBBIE system and
as well as the number of total entities, sentences include additional functionalities and features de-
and tokens. veloped to achieve its main objectives.
The annotated biomaterials corpus is The Biomaterials Annotator and the annotated
available and free for use; information corpus are open source and available to the com-
to access the corpus can be found at munity to promote future efforts in the field and
https://github.com/ProjectDebbie/ contribute towards its sustainability.
Biomaterials_annotator.
5 Conclusions and future directions 6 Acknowledgements
In this work we present the Biomaterials Annota-
tor, an ontology-based NER system that identifies This project has received funding from the Euro-
17 domain specific types of concepts and delivers pean Union Horizon 2020 programme under the
an annotated biomaterials corpus of 1222 MED- Marie Skodowska- Curie grant agreement DEB-
LINE articles available for future text mining and BIE, project number: 751277. O. H. is funded
machine learning efforts. We have carried out a val- through a Bosch-Aymerich fellowship. J-M.F, S.C-
idation activity to measure the performance of the G and J-L.P are partly supported by INB Grant
NER system, with the participation of nine bioma- (PT17/0009/0001 - ISCIII-SGEFI / ERDF). M-P. G.
terials experts, obtaining a global average F-score acknowledges the ICREA Academia Award from
of 0.77. Generalitat de Catalunya. J.C. is partly supported
Future work in the development of the system by eTRANSAFE (received funding from the Inno-
could involve annotating relations and linking iden- vative Medicines Initiative 2 Joint Undertaking un-
tified concepts to manufactured biomaterials ob- der grant agreement No 777365 and support from
jects. It may also include incorporating additional the European Union’s Horizon 2020 research and
categories using controlled resources. Improve- innovation programme and EFPIA).
ments to the system will continue in an iterative
43References Christopher Manning, Mihai Surdeanu, John Bauer,
Jenny Finkel, Steven Bethard, and David McClosky.
A. R. Aronson. 2001. Effective mapping of biomedical 2014. The Stanford CoreNLP Natural Language
text to the UMLS Metathesaurus: the MetaMap pro- Processing Toolkit. In Proceedings of 52nd An-
gram. Proceedings / AMIA ... Annual Symposium. nual Meeting of the Association for Computational
AMIA Symposium, pages 17–21. Linguistics: System Demonstrations, pages 55–60,
Stroudsburg, PA, USA. Association for Computa-
Olivier Bodenreider. 2004. The Unified Medical Lan- tional Linguistics.
guage System (UMLS): Integrating biomedical ter-
minology. Nucleic Acids Research. Marcos Martínez-Romero, Clement Jonquet, Martin J.
O’Connor, John Graybeal, Alejandro Pazos, and
David Campos, Sérgio Matos, and JoséLuís Oliveira. Mark A. Musen. 2017. NCBO Ontology Recom-
2013. Current Methodologies for Biomedical mender 2.0: An enhanced approach for biomedical
Named Entity Recognition. In Biological Knowl- ontology recommendation. Journal of Biomedical
edge Discovery Handbook, pages 839–868. John Wi- Semantics.
ley Sons, Inc.
Andrew McCallum and Wei Li. 2003. Early results for
Hyejin Cho and Hyunju Lee. 2019. Biomedical named entity recognition with conditional random
named entity recognition using deep neural net- fields, feature induction and web-enhanced lexicons.
works with contextual information. BMC Bioinfor- In Proceedings of the Seventh Conference on Natu-
matics, 20(1):735. ral Language Learning at HLT-NAACL 2003, pages
188–191.
Hamish Cunningham, Valentin Tablan, Angus Roberts,
and Kalina Bontcheva. 2013. Getting More Out of François Pognan, Thomas Steger-Hartmann, Car-
Biomedical Documents with GATE’s Full Lifecycle los Díaz, Niklas Blomberg, Frank Bringezu,
Open Source Text Analytics. PLoS Computational Katharine Briggs, Giulia Callegaro, Salvador
Biology. Capella-Gutierrez, Emilio Centeno, Javier Corvi,
Philip Drew, William C. Drewe, José M. Fernán-
Thanh Hai Dang, Hoang Quynh Le, Trang M. Nguyen, dez, Laura I. Furlong, Emre Guney, Jan A. Kors,
and Sinh T. Vu. 2018. D3NER: Biomedical named Miguel Angel Mayer, Manuel Pastor, Janet Piñero,
entity recognition using CRF-biLSTM improved Juan Manuel Ramírez-Anguita, Francesco Ronzano,
with fine-tuned embeddings of various linguistic in- Philip Rowell, Josep Saüch-Pitarch, Alfonso Valen-
formation. Bioinformatics, 34(20):3539–3546. cia, Bob van de Water, Johan van der Lei, Erik van
Mulligen, and Ferran Sanz. 2021. The etransafe
Zhou GuoDong and Su Jian. 2004. Exploring deep project on translational safety assessment through
knowledge resources in biomedical name recogni- integrative knowledge management: Achievements
tion. page 96. and perspectives. Pharmaceuticals, 14(3).
Osnat Hakimi, Josep Luis Gelpi, Martin Krallinger, Ariel S. Schwartz and Marti A. Hearst. 2003. A simple
Fabio Curi, Dmitry Repchevsky, and Maria-Pau algorithm for identifying abbreviation definitions in
Ginebra. 2020. The Devices, Experimental Scaf- biomedical text. Pacific Symposium on Biocomput-
folds, and Biomaterials Ontology (DEB): A Tool ing. Pacific Symposium on Biocomputing.
for Mapping, Annotation, and Analysis of Bio-
materials Data. Advanced Functional Materials, Hye Jeong Song, Byeong Cheol Jo, Chan Young Park,
30(16):1909910. Jong Dae Kim, and Yu Seop Kim. 2018. Com-
parison of named entity recognition methodologies
Jelena Jovanović and Ebrahim Bagheri. 2017. Seman- in biomedical documents. BioMedical Engineering
tic annotation in biomedicine: The current land- Online, 17(Suppl 2).
scape.
Yu Song, Eunju Kim, Gary Geunbae Lee, and Byoung-
Suwisa Kaewphan, Kai Hakala, Niko Miekka, Tapio kee Yi. 2004. POSBIOTM-NER in the shared task
Salakoski, and Filip Ginter. 2018. Wide-scope of BioNLP/NLPBA 2004.
biomedical named entity recognition and normaliza-
tion with CRFs, fuzzy matching and character level Hillel Taub-Tabib, Micah Shlain, Shoval Sadde, Dan
modeling. Database, 2018(2018):96. Lahav, Matan Eyal, Yaara Cohen, and Y. Goldberg.
2020. Interactive extractive search over biomedical
Jean Baptiste Lamy. 2017. Owlready: Ontology- corpora. ArXiv, abs/2006.04148.
oriented programming in Python with automatic
classification and high level constructs for biomed- Dennis G. Thomas, Rohit V. Pappu, and Nathan A.
ical ontologies. Artificial Intelligence in Medicine, Baker. 2011. NanoParticle Ontology for cancer nan-
80:11–28. otechnology research. Journal of Biomedical Infor-
matics.
Ki-Joong Lee, Young-Sook Hwang, and Hae-Chang
Rim. 2003. Two-phase biomedical NE recognition Federica Viti, Silvia Scaglione, Alessandro Orro, and
based on SVMs. Luciano Milanesi. 2014. Guidelines for managing
44data and processes in bone and cartilage tissue engi-
neering. BMC Bioinformatics, 15(Suppl 1):S14.
Chih Hsuan Wei, Hung Yu Kao, and Zhiyong Lu. 2013.
PubTator: a web-based text mining tool for assisting
biocuration. Nucleic acids research, 41(Web Server
issue).
Alexander Yeh, Alexander Morgan, Marc Colosimo,
and Lynette Hirschman. 2005. BioCreAtIvE task
1A: Gene mention finding evaluation. BMC Bioin-
formatics, 6(SUPPL.1):1–10.
Shaojun Zhao. 2004. Named entity recognition in
biomedical texts using an HMM model. (Grefen-
stette 1994):84.
45A Semantic resources
Table 3: List of semantic resources used by the Biomaterials Annotator
Semantic Resource Name Acronym Scope and relevance Type
Chemical Methods Methods used to collect chemical
1 CHMO Ontology
Ontology experiments data.
Chemical Entities of Compounds of biological relevance,
2 CHEBI Ontology
Biological Interest macromolecules.
The Devices, Experimental Biomaterials-related concepts,
3 Scaffolds and Biomaterials DEB materials, structures, Ontology
Ontology material processing.
EDAM Bioimaging EDAM- Imaging and sample preparation
4 Ontology
Ontology BIOIMAGING techniques.
Global Medical Device
5 GMDN Full names of medical devices. Nomenclature
Nomenclature
Biological concepts including
Interlinking ontology of biological phenomena, diseases,
6 IOBC Ontology
biological concepts molecular functions,
research imaging techniques.
A hierarchically organized
Controlled
7 Medical Subject Headings MeSH vocabulary produced by
vocabulary
the NLM.
National Cancer Institute Vocabulary for clinical care, Controlled
8 NCIT
Thesaurus translational and basic research. vocabulary
The description, preparation, and
9 Nanoparticle ontology NPO Ontology
characterization of nanomaterials.
Biomedical protocols, instruments,
Ontology for Biomedical
10 OBI data generated, materials, analysis Ontology
Investigations
performed.
Ontology of Nuclear Models, chemicals, tools, research
11 ONTOTOXNUC Ontology
Toxicity techniques and models.
Human disease terms, genomic,
Precision Medicine
12 PREMEDONTO molecular, phenotype, related Ontology
Ontology
medical vocabulary.
An integrated cross-species
13 Uber Anatomy Ontology UBERON Ontology
anatomy ontology
46B Annotation categories
Table 4: Annotation categories, their respective semantic resources and imported classes
Annotation
Definition Resource and imported classes
category
A non-drug raw material DEB: Biomaterials
1 Biomaterial or substance suitable for CHEBI: Macromolecule
inclusion in systems which MeSH: Biomedical or Dental Material
augment or replace the
function of bodily tissues
or organs.
Biomaterial Classification or nature
2 DEB: Biomaterial Type
Types of biomaterials.
Biologically Substance included in a DEB: Biologically Active Substance
3 Active manufactured object in MeSH: amino acid, peptide, protein
Substance order to impart a biological Biologically Active Substance
activity. Pharmacologic Substance
NCIT: Protein Domain
DEB: Manufactured Object
Manufactured A physical object created
4 MeSH: Medical device
Object by hand or machine.
GMDN: Full nomenclature
A part, region or
Manufactured
component referred to
5 Object DEB: Manufactured Object Component
as a distinct unit, such
Component
as a surface or a layer.
Medical Intended use, context, DEB: Medical Application
6
Application, function or outcome of MeSH: Disease or Syndrome
Disease the manufactured object. Therapeutic or Preventive Procedure
or condition Anatomical Abnormality
Manufactured Characteristics inherent DEB: Manufactured Object Features
7
Object or given during processing MeSH: Chemical Viewed Structurally
Features to a manufactured object or
its components.
The configuration, form
or texture associated with
8 Structure DEB: Structure
a manufactured object
or its components.
Associated A cellular or biological DEB: Associated Biological Process
9 Biological process that the MeSH: Organ or Tissue Function
Process manufactured object is Molecular Function
designed to cause Cell Function
or support, or is measured Biological function
to affect. NCIT: Cellular Process
Material A planned process which DEB: Material Processing
10
Processing results in physical changes CHMO: Material Processing
in a specified input
material.
The reported cell line MeSH: Cell
11 Cell or primary cell NCIT: Cell
type. UBERON: Bone cell, cardiocyte
circulating cell
connective tissue cell
epithelial cell
The species and /or
12 Species breed used in the MeSH: Mammal
study.
A tissue or an organ MeSH: Tissue,
13 Tissue mentioned in the Body Location or organ
study as the target Body part, organ or
or test system for organ component
the biomaterial object UBERON: Tissue
or medical device. PREMEDONTO:Body Part,
Organ, Organ System
47Table: Continued
Annotation
Definition Resource and imported classes
category
Adverse An unfavourable or DEB: Adverse Effects
14
Effects unintended disease, sign, MeSH: Pathologic Function
or symptom (including an
abnormal laboratory
finding) that is temporally
associated with the use
of a medical device or MeSH: Laboratory Procedure,
biomaterial. Molecular Biology Research Technique
The reported laboratory DEB: Research Technique
Research technique or instrument NCIT: Research Technique
15
Technique used in an experimental NPO: Instrument
study. IOBC: Microscope
OBI: Assay
EDAM: Imaging,
Sample preparation
ONTOTOXNUC: Outil
The effect associated with
Effect
manufactured object in
16 On Biological DEB: Effect On Biological System
a specific test system
System
(cells, tissue or organism).
The study set up,
Study
17 such as in vitro, DEB: Study Type
Type
in vivo, or clinical.
48You can also read
